pytorch
Background
-
Role: Full Stack Developer & Lead Developer of Synthesia & Avapre Projects
Expanded leadership scope beyond the Synthesia project to spearhead Avalia Preenche (Avapre), an advanced AI-powered document extraction system.
- Intelligent Pipelines: Engineered robust OCR and structured data extraction workflows using Pytorch, OpenCV and Large Language Models (LLMs), significantly reducing manual data entry for healthcare providers.
- Scalability: Optimized performance for high-volume processing in a critical healthcare environment.
- Legacy System Support: Identified and resolved critical bugs in their legacy platform, ensuring system stability and reliability while integrating new features.
-
A Deep Learning solution designed to detect forged SMTP headers. This project addresses the scarcity of labeled datasets in cybersecurity by employing Synthetic Data Generation to train robust models capable of identifying sophisticated phishing attempts.
- Tech Stack: Python, PyTorch.
- Outcome: Published as a full paper at SBSeg 2025.
-
Provider: Alura
Comprehensive training covering the end-to-end lifecycle of deep learning models. Focused on tensor operations, building custom neural architectures, and implementing training loops using PyTorch for complex non-linear problems.
-
Provider: Alura
I mastered the mechanics of deep model optimization by configuring advanced solvers like Adam and SGD, while strategically applying regularization methods such as Dropout and weight decay to ensure robust generalization and prevent overfitting during the backpropagation process.
-
Provider: Alura
Focused on sequential data architectures, I utilized PyTorch to design and implement Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU), effectively addressing the vanishing gradient problem to maintain temporal dependencies in complex time-series and natural language tasks.
-
Provider: Alura
Advanced implementation of computer vision models. Focused on CNN architecture design, including convolutional layers, pooling, and dropout strategies for feature extraction and image classification tasks.
-
Provider: Alura
This certification focused on the low-level implementation of neural architectures using Tensors and computational graphs, where I leveraged the torch.nn module to construct Multi-Layer Perceptrons and define custom weight initialization strategies for non-linear classification challenges.